OpenAI Dots features and capabilities

What “OpenAI Dots features” means here

A feature list is a promise if a vendor writes it, and a wish list if a community writes it. dotsbase is the second kind. The sections below describe jobs: give an agent a machine, let a team reach it in Slack, decide what email is allowed to do, and keep a person in the loop before something is sent or deleted. Where OpenAI has already documented a job, the paragraph says so and links the document. Where the job is only a common expectation of “Dots,” the paragraph says that too. The checklist at the bottom is your copy of the wish list. It is stored in this browser and it does not flip any switch on an OpenAI account.

If you only remember one distinction, remember this. OpenAI has published workspace agents that can live in ChatGPT and Slack, and has published hosted sandboxes that give a developer session a Linux computer. The phrase “OpenAI Dots features” in search usually wants both ideas plus a mailbox, under one product name. That combined product name is not what those two documents are titled. Read what is OpenAI Dots for the definition, then come back here for the jobs.

Dots agent VM

A Dots agent VM, in community talk, is a computer the agent can use while it works: a shell, a filesystem, and often a browser. The reason people ask for it is concrete. Some tasks are not a paragraph. They are a script, a spreadsheet, a page behind a login, a repository, a PDF that has to be filled from three other files. A chat box can describe those steps. A machine can perform them, and can also perform the wrong ones. The feature is not “intelligence.” The feature is a bounded computer with a log.

OpenAI’s published version of this idea for developers is the hosted sandbox in the Agents API. The docs describe a Linux workspace OpenAI provisions and connects. The working directory is on that machine. The application chooses network access: open, off, or limited to named hosts. Files can remain for the session, and outputs can be published as artifacts you download after the run. That is a real VM-shaped feature. It is requested by software you write. It is the citation to use when someone claims “OpenAI agents have no computer.” They have a documented computer for that API. It is not a citation that a ChatGPT consumer button called Dots already exposes the same machine.

Community expectations add a few clauses the API page does not have to satisfy. People want the machine to be reachable from a normal ChatGPT thread, not only from an application server. They want a browser that can be watched. They want the filesystem to feel like a project folder rather than a temporary scratch disk. They want the machine to survive a coffee break. Each of those clauses can be true of some OpenAI surface and false of another. Workspace agents are described as cloud agents that keep working when you step away. Sandboxes are described as session environments with their own lifetime. Dots, as a community name, blurs those lifetimes into “my agent’s computer.” When you evaluate a demo, ask which lifetime you are seeing.

What to demand from any agent computer

• A list of directories the agent may read, and a shorter list it may write.
• Network off, or network limited to hosts you named, unless the task truly needs the open web.
• A browser that does not silently reuse your personal cookies.
• Artifacts you can download, and a way to throw the machine away.
• A person in the loop before the machine talks to a system of record.

Those five lines are stricter than a keynote and looser than a security review. They are the right bar for a community guide. The VM task runbook in the template library, once you are signed in, is a brief written in that spirit. It is not an OpenAI policy.

Dots in Slack

Slack is the integration OpenAI has actually described for workspace agents. The help center walks through adding the agent to a channel, creating a Slack handle, and the condition that connectors use shared authentication. It warns that a personal login used as a shared connection can let other people trigger actions as you. That warning is the most important “feature” in the Slack story. The feature is not that a bot can talk. The feature is that the bot’s authority is explicit, shared on purpose, and limited to a channel you chose.

Community talk about Dots Slack often skips the admin step. People imagine mentioning an agent the way they mention a coworker, and getting a file back. The published path is pickier: a workspace that is allowed to use agents, a Slack app an admin approves, shared connections, and a channel. If your question is “can OpenAI’s documented agent sit in Slack,” the answer in OpenAI’s own preview is yes, for the plans and roles that preview includes. If your question is “can every ChatGPT user do that today under the name Dots,” this site does not have a public document that says yes.

Dots and email

Email is the other half of the phrase “Dots Slack email,” and it is on a different footing. People want triage, drafts, and follow-ups. Muse’s public story includes mail as a life-admin connector. OpenAI’s workspace-agent introduction emphasizes ChatGPT and Slack, with other surfaces described as coming later in that announcement’s framing. So email belongs in the community expectation for Dots, and it belongs in your checklist if you care about it. It does not belong in a sentence that claims a documented Dots inbox. Drafting is a different permission from sending. A useful brief allows the draft and forbids the send until a person presses a button. Several templates on this site are written that way.

Schedules beside Slack and email

Schedules sit beside both channels. Workspace agents can be set to run when you are not in the thread. A morning digest in Slack, or a pile of drafts waiting in the morning, is the job people mean. A schedule without a scope is just a bot that can surprise you. Pair every schedule with a destination, a source list, and a rule for what it must not do unattended. That pairing is a feature even when the vendor UI is still a text box.

Sharing, approvals, and limits for a Dots agent

The rest of the feature set is control. Sharing means one agent definition used by a team, with an owner and a way to disable it. OpenAI’s workspace-agent material is explicitly about that kind of sharing inside an organization. A personal Dots conversation sometimes wants the opposite: an agent that is only yours. Those are both coherent products. They should not be described as one toggle.

Approvals are how a computer-using agent stays employable. Sending mail, paying, deleting, pushing code, and posting to a channel are side effects. A community guide that lists them as “supported integrations” without a stop is advertising. Put the stop in the brief. The checklist includes it as a preference you can record before you are offered a settings screen.

Limits are the feature people forget until the invoice. A VM and a browser burn more compute than a short chat. Any honest product will meter them. The pricing page refuses to invent the meter. It does say that a Pro-like allowance is a community guess for heavy computer use, and that ChatGPT’s own Pro prices are not a Dots price. When you compare Dots vs Muse or Dots vs Grok Bot, compare the limit story too: whose quota, whose admin, whose log.

Your OpenAI Dots feature checklist

Tick what you want. The boxes are written as preferences — “a cloud computer I can point at a task” — not as claims that OpenAI has flipped them on for your account. The state is saved under a localStorage key in this browser. Clearing site data clears the list. Nothing is sent to dotsbase, and nothing is sent to OpenAI. If you use the Chinese pages, the same browser store is used, so the ticks follow you across languages.

When a public Dots client exists, redo the list against its settings page. Until then, the list is a way to keep a meeting honest. For the aliases people mix into feature requests — Aeon, orbit, O — use the names guide. For starter briefs that encode the stops above, use templates after GitHub sign-in. The Chinese feature guide is OpenAI Dots 功能.

Published pages for the documented half

• Introducing workspace agents in ChatGPT — cloud agents, Slack, schedules, sharing.
• Workspace agents help — channels, handles, shared auth, admin notes.
• OpenAI-hosted sandboxes — the developer VM: Linux, files, network policy, artifacts.